26% of the sentences tripped the detection threshold for their classifier. That isn't a detection probability on the entire text.